The influences of the spatial extent selection for non-landslide samples on statistical-based landslide susceptibility modelling: a case study of Anhui Province in China

Abstract

Landslide susceptibility assessment is crucial to the development of appropriate strategies to mitigate the risk of landslide fatalities and economic losses. The selection of spatial extent for non-landslide samples has an important role in the statistical-based landslide susceptibility modelling (LSM). In this study, two different non-landslides sampling areas (the entire area and the mountainous area of ​​Anhui Province, China) were designed to explore the influences of the different spatial extent for non-landslides sampling on LSM. Six categories of influencing factors including climatic, morphological, geological, hydrological, vegetation, and human activities were considered. The dominant influencing factors that are more closely related to the distribution of historical landslides were selected based on the GeoDetector. The landslide inventory samples and the non-landslide samples generated on two selected areas were divided into a training set (70%) and a validation set (30%) for establishing the entire area LR model (EaeraLR) and the mountainous area LR model (MareaLR) based on the logistic regression (LR) model. The performance of the models was evaluated by the confusion matrix and the area under the receiver operating characteristic curve (AUROC). The results showed that the EareaLR model outperformed the MareaLR model by various evaluation metrics and the appearance of the final landslide susceptibility map. Hence, we conclude that the potential influence of the spatial extent of the non-landslide sample selection needs to be taken into account while comparing the reliability of different data-driven landslide susceptibility models.